1. Field of the Invention
The present invention relates to microelectromechanical system (MEMS) resonators and filters, and, more particularly, to the fabrication of such devices in a manner which allows integration with other integrated circuit technologies, such as Bi-CMOS, while maintaining the desired properties of these devices such as high resonant frequency (f0) and very high quality factor (Q).
2. Description of the Prior Art
Microelectromechanical system (MEMS) devices have the potential for great impact on the communications industry. MEMS RF switches, oscillators (resonators), filters, varactors, and inductors are a few of the devices that could replace large and relatively expensive off-chip passive components. It is even possible that the introduction of these types of MEMS devices, particularly resonators and filters, into analog and mixed-signal integrated circuits could dramatically alter the architecture of current wireless communication devices. Key to such advancements is the ability to monolithically integrate MEMS RF components with integrated circuit technologies to realize cost, size, power, and performance benefits.
MEMS resonators and filters have been under development for some time. For resonators and filters aimed at RF communications applications, the key design factors are ability to reach the frequencies of interest (approx. 900 MHz-2 GHz), low voltage operation, small size, and very high quality factor (Q). Resonators and filters developed to date have demonstrated high Q's and reasonably small sizes, but have not achieved the frequency or bias voltage targets required for incorporation with analog and mixed signal circuits. Other drawbacks of current MEMS resonators and filters include incompatibility of materials, processes, and processing temperatures for integration with other IC processes, inability to scale the devices to the desired sizes because of grain size limitations of the materials used and inability to form very small gaps between electrodes, and failure to provide protection for the MEMS devices from subsequent processing steps and ambient conditions and contamination.
Typical designs of prior art MEMS resonators and filters are illustrated in FIGS. 1A-1C. FIG. 1A shows a comb-drive type MEMS filter. Stationary combs 1 and 7 are connected via anchors 2 and 8 to input and output electrodes 3 and 9, respectively. Moving comb 4 is connected via anchors 5 to ground plane 6. The fingers of all three comb structures are suspended above the underlying substrate, the ground plane, and the input and output electrodes except at the anchor points. All three combs are comprised of a conductive material, typically heavily doped polysilicon. The ground plane and input and output electrodes are also conductors typically made from heavily doped polysilicon. During operation, ground plane 6 is electrically contacted to the ground potential. The potential of moving comb 4 is also at ground. An AC excitation, superimposed on a DC bias, is applied to input electrode 3 and thus, via anchor 2, to stationary comb 1. The same DC bias is applied to output electrode 9 and thus, via anchor 8, to stationary comb 7. Because of the potential difference between the fingers of stationary comb 1 and moving comb 4, moving comb 4 is attracted laterally toward stationary comb 1. The magnitude of this potential difference, and thus the distance which moving comb 4 travels, is modulated by the AC excitation. When the frequency of the exciting AC voltage closely matches the mechanical resonant frequency f0 of moving comb 4, the amplitude of vibration of moving comb 4 reaches a maximum that is dependent on the quality factor Q of the system. Simultaneously, the fingers of moving comb 4 and stationary comb 7 comprise a time-varying capacitor as the amount of overlap between the fingers of the combs changes with the movement of moving comb 4. Thus, through the relationship I=d(CV)/dt, there will also be a time-varying current which can be sensed electrically at output electrode 9. The magnitude of this current will also be greatest when the frequency of the exciting AC voltage at input electrode 3 closely matches the f0. Thus, the device provides electromechanical filtering of the input signal around f0.
FIG. 1B shows another example of a prior art MEMS filter which is aimed at achieving higher-frequency operation. Two beams 11 and 15 are connected to ground electrodes 13 via anchors 12. Beams 11 and 15 are also connected to one another by bridge 14. Taken alone, either beam 1 for beam 15 comprises a MEMS resonator. Coupling two or more MEMS resonators together creates the MEMS filter. Beams 11, 15, and bridge 14 are suspended above the underlying substrate. Ground electrodes 13, and input and output electrodes 16 and 17 (respectively) are also suspended above the underlying substrate except at anchor points 12. Beams 11 and 15, bridge 14, and all electrodes 13, 16 and 17 are composed of a conductive material, typically heavily doped polysilicon. During operation, ground electrodes 13 are electrically contacted to the ground potential; thus, via anchors 12, the potential of beams 11 and 15 and bridge 14 are also at ground. An AC excitation, superimposed on a DC bias, is applied to input electrode 16. The same DC bias is applied to output electrode 17. Because of the potential difference between them, beam 11 is attracted downward toward electrode 16. The magnitude of this potential difference, and thus the distance which beam 11 travels, is modulated by the AC excitation. When the frequency of the exciting AC voltage closely matches the mechanical resonant frequency f0 of beam 11, the amplitude of vibration of beam 11 reaches a maximum that is dependent on the quality factor Q of the system. The mechanical energy of vibration of beam 11 is transmitted via bridge 14 to beam 15. Beam 15 and output electrode 17 comprise a time-varying capacitor as the distance between the two structures changes with the movement of beam 15. Thus, through the relationship I=d(CV)/dt, there will also be a time-varying current which can be sensed electrically at output electrode 17. The magnitude of this current will also be greatest when the frequency of the exciting AC voltage at input electrode 16 closely matches the f0. Thus, the device provides electromechanical filtering of the input signal around f0.
FIG. 1C shows cross section A-A′ of prior art MEMS resonator 11 as seen in FIG. 1B. This cross section also shows the substrate 21 upon which the MEMS resonator or filter is constructed. This substrate is typically silicon (Si), although other substrates such as glass, quartz, or gallium arsenide (GaAs) have also been used. Also shown is insulating layer 22, typically silicon dioxide (SiO2), used to electrically isolate the MEMS device from the substrate and other devices. Air gap 23 can be seen in the cross section, demonstrating that beam 11 is freestanding except at anchor points 12. Not shown here is the sacrificial material that occupied gap 23 during the construction of this device, and was later removed so that beam 11 would be free to vibrate.
One of the drawbacks of the prior art is the deposition temperature of the materials commonly used for construction of the MEMS device. Although various conductive materials have been used to form MEMS resonators and filters, polysilicon is the most common. Polysilicon is frequently chosen because of its relatively high ratio of elastic modulus (E) to density (ρ). This ratio is one of the most important factors in determining the resonant frequency of the device, and since high frequencies are sought for RF communications applications, high ratios of E/ρ are desirable. However, polysilicon must be deposited at temperatures in excess of 600° C. Furthermore, the dopant atoms, such as phosphorus, which are added to the polysilicon to make it sufficiently conductive, frequently must be annealed at temperatures near 900° C. in order to activate them. These temperatures are well above the temperatures used in fabrication of the metal interconnect levels of integrated circuit processes. This means that prior art MEMS resonators and filters, if they were to be integrated in an IC process, would have to be fabricated at the same time as the transistor devices (which permit higher processing temperatures). This type of process integration is much more difficult to achieve and is very specific to the particular IC process. Thus, the process steps for formation of the MEMS device would likely need to be altered each time there was a change to the IC process, or whenever it was desired to integrate the MEMS device with a different IC process. A much simpler and more modular approach is to integrate the MEMS device after all circuit processing, including interconnect levels, has been completed. However, this cannot be done with prior art MEMS resonators and filters.
Another serious issue with prior art MEMS resonators and filters is the process by which the devices are released from the surrounding layers and substrate. The most commonly used sacrificial material (i.e., the material which temporarily occupies the gap region and is later removed to create the freestanding MEMS structure) in the prior art is SiO2. This material is removed by means of etching in an aqueous buffered hydrofluoric acid (buffer-HF) solution. This solution will also remove silicon nitride (SiN), although at a slower rate, and causes etching of or damage too many metals. Because SiO2 and SiN are used as insulating layers in integrated circuits, this release method also makes it very difficult to integrate prior art MEMS resonators and filters with IC processes. Another problem with the use of aqueous buffer-HF as a release method is the occurrence of a phenomenon known as stiction. After the sacrificial SiO2 has been removed, the buffer-HF is rinsed away. As the water is then removed during the subsequent drying step, the freestanding MEMS parts have a tendency to stick to the substrate or surrounding materials because of the high surface tension of the water. Prior art MEMS devices frequently have to be subjected to an alternative drying method such as the use of supercritical carbon dioxide (CO2). This method and the associated tools are also not part of any current IC process flow. Another drawback to using aqueous buffer-HF to remove the sacrificial layer is that it restricts the aspect ratios and gap dimensions that can be achieved in MEMS devices. Very small gaps (tens—few hundred nanometers) cannot be formed because of limited transport of the etchant and etch products in and out of the gap region. Small gaps are desirable in MEMS devices because they allow the use of lower actuation voltages. Typical RF IC's use supply voltages of 3V; most prior art MEMS resonators and filters require biases of 20V and up.
Another concern with prior art MEMS resonators and filters is the lack of adequate encapsulation of the devices for protection during subsequent processing steps, and from ambient contamination, humidity, and pressure when fabrication is complete. Once the MEMS device has been released, additional processing steps create the risk of re-filling the gap area with deposited material and re-connecting the device to the substrate, causing failures due to stiction, or adversely affecting yield or performance via the introduction of particulates to the gap region or the device itself. Even after all fabrication is complete, MEMS resonators and filters are quite sensitive to ambient conditions. For example, a particulate adhering to the resonator beam could change the mass (and thus the resonant frequency) of a small beam by several hundred percent. A particulate lodged in the gap region would damp or completely prevent resonance. Finally, it has been established that the quality factor of MEMS resonators and filters is directly related to ambient pressure, and in order to maximize Q, MEMS resonators and filters must be operated at pressures below about 0.1 Torr. Several encapsulation schemes have been proposed in the prior art. The most common methods involve bonding a second substrate with an etched cavity over the MEMS device by various means (e.g. anodic bonding, eutectic bonding, etc.). However, to date these methods have not been adequately demonstrated at wafer scale. Each individual device must be capped. This method is not compatible with reasonable manufacturing processes. This method also causes difficulties with integrated circuit designs intended for packaging via flip-chip (solder bump) die attach. Furthermore, this method assumes that the MEMS resonator or filter is the last device fabricated (i.e., it is exposed on the top surface of the chip), and it has already been seen that prior-art MEMS resonators and filters are not compatible with fabrication after the completion of IC processing. Another encapsulation method that has been proposed in the prior art is to cover the MEMS resonator with additional SiO2, then to cap the entire structure with a shell of porous polysilicon. The device is then exposed again to aqueous buffer-HF, which is transported through the porous polysilicon, removes the covering SiO2, and diffuses back out through the porous polysilicon. This method is unsatisfactory for many reasons, several of which (deposition temperature of polysilicon and stiction) have already been discussed.